Data Monitor Agent

A configurable agent that watches metrics you care about — API response times, database row counts, Stripe revenue — and alerts you when thresholds are crossed.

Role Description

The Data Monitor Agent polls configured data sources on a schedule, compares current values against defined thresholds, and sends alert emails with AI-generated context when anomalies are detected.

Difficulty: Intermediate
Build time: ~1 hour to configure data sources
Stack: TypeScript, Anthropic SDK, Supabase, Resend

When to Use

Use this agent when you:

• Check dashboards manually multiple times a day
• Want to be notified when a metric drops below a threshold (signups, revenue, uptime)
• Need to monitor metrics across multiple systems without building a full observability stack
• Want AI-generated context on why a metric changed, not just that it changed

How It Works

Trigger → Context → Intelligence → Action:

1. Trigger: Cron schedule (configurable interval, minimum 1 minute)
2. Context: Queries configured data sources — APIs, Supabase, Stripe
3. Intelligence: Claude compares current values to thresholds, generates context on why a metric may have changed
4. Action: Sends an alert email with the anomaly and AI-generated explanation

System Prompt / Key Instructions

You are a data monitoring analyst. You will receive:
1. A set of metrics with their current values
2. Their historical context (last 7 days of data)
3. Their configured thresholds

Your job:
1. Identify which metrics have crossed their alert thresholds
2. For each alert, write a 2-3 sentence explanation:
   - What happened (the specific anomaly)
   - Likely cause (based on historical patterns and context)
   - Recommended action (what to check or do)
3. Assign a severity: LOW | MEDIUM | HIGH | CRITICAL
4. Return structured JSON

Only alert on real anomalies. Don't alert if a metric is within normal variance.
Use the historical data to distinguish "this is just normal fluctuation" from 
"this is actually unusual."

Key Configuration

// monitors.ts
export const MONITORS = [
  {
    id: 'daily-signups',
    name: 'Daily Signups',
    description: 'New user registrations in the last 24 hours',
    query: async () => {
      const { count } = await supabase
        .from('users')
        .select('*', { count: 'exact', head: true })
        .gte('created_at', new Date(Date.now() - 86400000).toISOString());
      return count ?? 0;
    },
    thresholds: {
      critical: { below: 5 },   // Alert if fewer than 5 signups/day
      warning: { below: 15 },   // Warning if fewer than 15
    },
    unit: 'signups',
  },
  {
    id: 'stripe-mrr',
    name: 'Active Subscriptions',
    description: 'Current count of active Stripe subscriptions',
    query: async () => {
      const subscriptions = await stripe.subscriptions.list({ status: 'active' });
      return subscriptions.data.length;
    },
    thresholds: {
      critical: { below: 10, changePercent: -20 }, // Alert if drops 20%
    },
    unit: 'subscriptions',
  },
  {
    id: 'api-latency',
    name: 'API P95 Latency',
    description: 'P95 response time for critical API endpoints',
    query: async () => {
      // Query your logging service
      return await getP95Latency('/api/v1/');
    },
    thresholds: {
      warning: { above: 500 },   // 500ms warning
      critical: { above: 2000 }, // 2s critical
    },
    unit: 'ms',
  },
];

Safeguards Built In

The template includes a minimum polling interval safeguard:

// Prevents running more frequently than once per minute
const MIN_INTERVAL_MS = 60_000;

// Cooldown period — won't re-alert the same metric within 4 hours
const ALERT_COOLDOWN_MS = 4 * 60 * 60 * 1000;

// Historical data tracked to distinguish noise from anomalies
const HISTORY_DAYS = 7;

Example Usage

npm install
cp .env.example .env
# Configure: ANTHROPIC_API_KEY, SUPABASE_URL, SUPABASE_SERVICE_KEY, RESEND_API_KEY

# Test run (checks thresholds once)
npm start

# Run on schedule (local cron)
# Add to crontab: */15 * * * * cd /path/to/monitor && npm start

# Deploy to Vercel with cron
vercel deploy

Adding a new monitor:

1. Add a new object to MONITORS in monitors.ts
2. Define the query function (async, returns a number)
3. Set thresholds for warning and critical levels
4. Specify unit for readable alerts

Features

• Configurable metric sources (API, Supabase database, Stripe)
• Threshold-based alerting with cooldown periods
• AI-generated context for why a metric changed
• Minimum polling interval safeguard (prevents API abuse)
• Historical trend tracking (7-day rolling window)
• Multi-severity levels: LOW, MEDIUM, HIGH, CRITICAL